[Standup] Add all rendered environment variables as a configmap

scenarios/examples/cpu.sh

@@ -4,14 +4,17 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
-#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
 # PVC parameters
 #             Storage class (leave uncommented to automatically detect the "default" storage class)
@@ -24,6 +27,8 @@
 #export LLMDBENCH_DEPLOY_METHODS=standalone
 #export LLMDBENCH_DEPLOY_METHODS=modelservice
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 export LLMDBENCH_VLLM_COMMON_TENSOR_PARALLELISM=2
 export LLMDBENCH_VLLM_COMMON_AFFINITY=kubernetes.io/os:linux
 export LLMDBENCH_VLLM_COMMON_ACCELERATOR_NR=0

scenarios/examples/gpu.sh

@@ -4,6 +4,7 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
@@ -13,6 +14,7 @@
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
 # PVC parameters
 #             Storage class (leave uncommented to automatically detect the "default" storage class)
@@ -26,6 +28,9 @@
 ######export LLMDBENCH_DEPLOY_METHODS=standalone
 #export LLMDBENCH_DEPLOY_METHODS=modelservice
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
+
 #             Affinity to select node with appropriate accelerator (leave uncommented to automatically detect GPU... WILL WORK FOR OpenShift, Kubernetes and GKE)
 #export LLMDBENCH_VLLM_COMMON_AFFINITY=nvidia.com/gpu.product:NVIDIA-H100-80GB-HBM3        # OpenShift
 #export LLMDBENCH_VLLM_COMMON_AFFINITY=gpu.nvidia.com/model:H200                           # Kubernetes
@@ -119,6 +124,7 @@ vllm serve REPLACE_ENV_LLMDBENCH_DEPLOY_CURRENT_MODEL \
 EOF
 
 # llm-d Parameters
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=kgateway
 export LLMDBENCH_VLLM_MODELSERVICE_PREFILL_TENSOR_PARALLELISM=$LLMDBENCH_VLLM_COMMON_TENSOR_PARALLELISM # (default is "1")
 #export LLMDBENCH_VLLM_MODELSERVICE_PREFILL_REPLICAS=1 # (default is "1")
 #export LLMDBENCH_VLLM_MODELSERVICE_PREFILL_EXTRA_VOLUME_MOUNTS=${LLMDBENCH_VLLM_COMMON_EXTRA_VOLUME_MOUNTS}

scenarios/examples/sim.sh

@@ -4,19 +4,24 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
-#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
 # Deploy methods
 #export LLMDBENCH_DEPLOY_METHODS=standalone
 #export LLMDBENCH_DEPLOY_METHODS=modelservice
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 export LLMDBENCH_VLLM_COMMON_TENSOR_PARALLELISM=0
 export LLMDBENCH_VLLM_COMMON_AFFINITY=kubernetes.io/os:linux
 export LLMDBENCH_VLLM_COMMON_ACCELERATOR_NR=0
@@ -60,4 +65,4 @@ export LLMDBENCH_HARNESS_NAME=inference-perf # (default is "inference-perf")
 #export LLMDBENCH_HARNESS_NAME=vllm-benchmark
 
 #export LLMDBENCH_HARNESS_EXPERIMENT_PROFILE=sanity_random.yaml # (default is "sanity_random.yaml")
-######export LLMDBENCH_HARNESS_EXPERIMENT_PROFILE=nop.yaml
+######export LLMDBENCH_HARNESS_EXPERIMENT_PROFILE=nop.yaml

scenarios/examples/spyre.sh

@@ -4,15 +4,17 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
-#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
 #export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
 export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
-#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
 # PVC parameters
 #             Storage class (leave uncommented to automatically detect the "default" storage class)
@@ -26,6 +28,8 @@ export LLMDBENCH_VLLM_COMMON_EXTRA_PVC_NAME=spyre-precompiled-model
 #export LLMDBENCH_DEPLOY_METHODS=standalone
 #export LLMDBENCH_DEPLOY_METHODS=modelservice
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 export LLMDBENCH_VLLM_COMMON_ACCELERATOR_RESOURCE=ibm.com/spyre_vf
 export LLMDBENCH_VLLM_COMMON_TENSOR_PARALLELISM=2
 export LLMDBENCH_VLLM_COMMON_AFFINITY="ibm.com/spyre.product:IBM_Spyre"
@@ -148,7 +152,7 @@ export LLMDBENCH_VLLM_MODELSERVICE_PREFILL_ACCELERATOR_RESOURCE=$LLMDBENCH_VLLM_
 
 # Decode parameters: 2 decode pods
 export LLMDBENCH_VLLM_MODELSERVICE_DECODE_REPLICAS=1
-export LLMDBENCH_VLLM_MODELSERVICE_DECODE_TENSOR_PARALLELISM=4
+export LLMDBENCH_VLLM_MODELSERVICE_DECODE_TENSOR_PARALLELISM=${LLMDBENCH_VLLM_COMMON_TENSOR_PARALLELISM}
 export LLMDBENCH_VLLM_MODELSERVICE_DECODE_ACCELERATOR_RESOURCE=$LLMDBENCH_VLLM_COMMON_ACCELERATOR_RESOURCE
 export LLMDBENCH_VLLM_MODELSERVICE_DECODE_CPU_NR=$LLMDBENCH_VLLM_COMMON_CPU_NR
 export LLMDBENCH_VLLM_MODELSERVICE_DECODE_CPU_MEM=$LLMDBENCH_VLLM_COMMON_CPU_MEM

scenarios/guides/inference-scheduling.sh

@@ -10,6 +10,13 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
@@ -22,6 +29,8 @@
 #export LLMDBENCH_VLLM_COMMON_PVC_STORAGE_CLASS=ocs-storagecluster-cephfs
 export LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE=1Ti
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 # Routing configuration (via gaie)
 #export LLMDBENCH_VLLM_MODELSERVICE_GAIE_PLUGINS_CONFIGFILE="default-plugins.yaml" # already the default
 

scenarios/guides/pd-disaggregation.sh

@@ -10,9 +10,15 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
-#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
@@ -23,6 +29,8 @@
 #export LLMDBENCH_VLLM_COMMON_PVC_STORAGE_CLASS=ocs-storagecluster-cephfs
 export LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE=1Ti
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 # Routing configuration (via gaie)
 #export LLMDBENCH_VLLM_MODELSERVICE_GAIE_PLUGINS_CONFIGFILE="default-plugins.yaml" # already the default
 

scenarios/guides/precise-prefix-cache-aware.sh

@@ -10,8 +10,15 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
-export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
@@ -22,9 +29,58 @@ export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_VLLM_COMMON_PVC_STORAGE_CLASS=ocs-storagecluster-cephfs
 export LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE=1Ti
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 # Routing configuration (via gaie)
+export LLMDBENCH_LLMD_INFERENCESCHEDULER_IMAGE_TAG=v0.4.0
 #export LLMDBENCH_VLLM_MODELSERVICE_GAIE_PLUGINS_CONFIGFILE="default-plugins.yaml" (default is "plugins-v2.yaml")
-export LLMDBENCH_VLLM_MODELSERVICE_GAIE_PLUGINS_CONFIGFILE="precise-prefix-cache-aware"
+export LLMDBENCH_VLLM_MODELSERVICE_GAIE_PLUGINS_CONFIGFILE="precise-prefix-cache-config.yaml"
+export LLMDBENCH_VLLM_MODELSERVICE_GAIE_CUSTOM_PLUGINS=$(mktemp)
+cat << EOF > $LLMDBENCH_VLLM_MODELSERVICE_GAIE_CUSTOM_PLUGINS
+precise-prefix-cache-config.yaml: |
+  apiVersion: inference.networking.x-k8s.io/v1alpha1
+  kind: EndpointPickerConfig
+  plugins:
+    - type: single-profile-handler
+    - type: precise-prefix-cache-scorer
+      parameters:
+        indexerConfig:
+          tokenProcessorConfig:
+            blockSize: 64
+            hashSeed: "42"
+          tokenizersPoolConfig:
+            hf:
+              tokenizersCacheDir: "/tmp/tokenizers"
+    - type: kv-cache-utilization-scorer
+    - type: queue-scorer
+    - type: max-score-picker
+  schedulingProfiles:
+    - name: default
+      plugins:
+        - pluginRef: precise-prefix-cache-scorer
+          weight: 3.0
+        - pluginRef: kv-cache-utilization-scorer
+          weight: 2.0
+        - pluginRef: queue-scorer
+          weight: 2.0
+        - pluginRef: max-score-picker
+EOF
+export LLMDBENCH_VLLM_MODELSERVICE_INFERENCE_POOL_PROVIDER_CONFIG=$(mktemp)
+cat << EOF > $LLMDBENCH_VLLM_MODELSERVICE_INFERENCE_POOL_PROVIDER_CONFIG
+destinationRule:
+  host: REPLACE_ENV_LLMDBENCH_DEPLOY_CURRENT_MODEL_ID_LABEL-gaie-epp
+  trafficPolicy:
+    connectionPool:
+      http:
+        http1MaxPendingRequests: 256000
+        maxRequestsPerConnection: 256000
+        http2MaxRequests: 256000
+        idleTimeout: "900s"
+      tcp:
+        maxConnections: 256000
+        maxConnectionDuration: "1800s"
+        connectTimeout: "900s"
+EOF
 
 # Routing configuration (via modelservice)
 #export LLMDBENCH_VLLM_MODELSERVICE_INFERENCE_MODEL=true # already the default

scenarios/guides/simulated-accelerators.sh

@@ -8,6 +8,8 @@ export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 export LLMDBENCH_VLLM_COMMON_REPLICAS=1
 
 export LLMDBENCH_VLLM_COMMON_ACCELERATOR_NR=0

scenarios/guides/tiered-prefix-cache.sh

@@ -10,8 +10,15 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
-export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
@@ -22,6 +29,8 @@ export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
 #export LLMDBENCH_VLLM_COMMON_PVC_STORAGE_CLASS=ocs-storagecluster-cephfs
 export LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE=1Ti
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 # Routing configuration (via gaie)
 #export LLMDBENCH_VLLM_MODELSERVICE_GAIE_PLUGINS_CONFIGFILE="default-plugins.yaml" # already the default
 

scenarios/guides/wide-ep-lws.sh

@@ -10,10 +10,17 @@
 
 # Model parameters
 #export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-0.6B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="Qwen/Qwen3-32B"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="RedHatAI/Llama-3.3-70B-Instruct-FP8-dynamic"
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-vision-3.3-2b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-speech-3.3-8b
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-8b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-granite/granite-3.3-2b-instruct
+#export LLMDBENCH_DEPLOY_MODEL_LIST=ibm-ai-platform/micro-g3.3-8b-instruct-1b
 #export LLMDBENCH_DEPLOY_MODEL_LIST="facebook/opt-125m"
 #export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-8B-Instruct"
-export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
-#export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
+#export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
+export LLMDBENCH_DEPLOY_MODEL_LIST="deepseek-ai/DeepSeek-R1-0528"
 
 # PVC parameters
 #             Storage class (leave uncommented to automatically detect the "default" storage class)
@@ -22,6 +29,8 @@ export LLMDBENCH_DEPLOY_MODEL_LIST="meta-llama/Llama-3.1-70B-Instruct"
 #export LLMDBENCH_VLLM_COMMON_PVC_STORAGE_CLASS=ocs-storagecluster-cephfs
 export LLMDBENCH_VLLM_COMMON_PVC_MODEL_CACHE_SIZE=2Ti
 
+#export LLMDBENCH_VLLM_MODELSERVICE_GATEWAY_CLASS_NAME=istio
+
 # Routing configuration (via gaie)
 export LLMDBENCH_VLLM_MODELSERVICE_GAIE_PLUGINS_CONFIGFILE="custom-plugins.yaml"
 export LLMDBENCH_VLLM_MODELSERVICE_GAIE_CUSTOM_PLUGINS=$(mktemp)